INFINITE SYSTEMS OF NONLINEAR OSCILLATION EQUATIONS RELATED TO THE STRING1 R. W. DICKEY 1. Introduction. The purpose of this paper is to discuss the exis- tence and uniqueness of solutions to an infinite system of nonlinear oscillation equations of the form (1.1) T'J + / (a0 + axf: tlti Tj=0, j = 1, 2, • • • , °o, where the constants a<>and ax satisfy the conditions a0^0 and ai>0 (the prime in (1.1) indicates differentiation with respect to /). The initial conditions on (1.1) will be taken as (1.2a) r,(0) = ay, (L2b) Tj(0) = 0j. Equations of the type (1.1) are related to the Duffing equation (cf. [l]), and arise in attempting to find Fourier series solutions oo (1.3) W(x, t) = X) Tj(t) sinJTrx/L, 3-1 to the nonlinear partial integro-differential equation (1.4) Wtt-(lIo + Hxf Wi(t,t)2di\wxx = 0, (Ho^O, Hx>0). Equation (1.4) describes the small amplitude vibra- tions of a string in which the dependence of the tension on the defor- mation cannot be neglected (cf. [2], [3]). The equations (1.1) form an infinite Hamiltonian system, and in fact there is no difficulty in showing that any solution of (1.1) satisfies the condition (1-5) ~( ± (r'j)2+ ao±jYj +1 ( Z/r:)2) = 0. Received by the editors May 12, 1969. 1 This research was sponsored by the National Science Foundation, Contract No. GP-7543. 459 License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use 460 R. W. DICKEY [December At first glance it would appear that if the initial conditions (1.2) satisfy a finite energy condition, i.e., OO 00 ~ / 00 \ 2 (1.6) h = £ ft + a0Z/«;2 + V ( Ufa*) < oo, j-l /-I 2 \ y_j / then (1.1) should have a solution for all t>0. Indeed this is the case for finite systems of the form (1.1) since the finite system (1.7) f/ +f (a0 + ax £ flti Tj = 0, j=l,2,---,N, has associated with it a Lipschitz constant (depending on N). Thus the method of successive approximation (cf. [4]) may be used to show the existence of a solution to (1.7) locally, and the continuation of this solution is guaranteed by the fact that the energy—and hence the solution and its derivative—remains bounded. However, the infinite system of equations (1.1) is not Lipschitz continuous because of the unbounded nature of the coefficient of Tj as j—>oo. Thus the method of successive approximation fails and an alternative proce- dure is necessary. In §2 of this paper, it will be shown that under certain conditions on the initial data (1.2), solutions of the finite system (1.7) converge to a solution of (1.1) as A7—>oo.In order to guarantee this it will be necessary to require that the initial data (1.2) satisfies a condition stronger than the simple finite energy condition (1.6). In §3 it will be shown that the solution of (1.1) satisfying the initial conditions (1.2) is unique among a certain class of functions. 2. Existence. In proving the existence of solutions to (1.1), it is convenient to define a set of functions Tj,n as follows: iorj^N, Tj,N is to be a solution of the finite system of equations (1.7) and satisfy the initial conditions (1.2) for j=l, 2, • • • , N, and for j>N set Tj,n —0. The functions Tj,n are of course also solutions of the infinite system (1.1), i.e. (2.1) T"N +fANTj,N = 0, j = 1, 2, • • • , oo, where 00 (2.2) Atr = ao + ax2-,l Ti,N. t~i If in addition the initial data (1.2) satisfies the finite energy condition (1.6) it follows that (cf. (1.5)) License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use 1969] NONLINEAROSCILLATION EQUATIONS 461 (2.3) ± (T'j.*)2+ aQ±/tIn + -£ ( tfti*)' ^ h. i-l j-1 2 \ J=i / Thus there exist constants Mx and M2 independent of N such that (2.4a) £ (Tj.n)* < Mx, 00 (2.4b) £/P/U < Mt. y-i It is a consequence of (2.4b) that the functions An are uniformly bounded independent of N. If it could also be shown that | A'N\ was uniformly bounded independent of N—so that the sequence {An} is not only bounded but equicontinuous—the existence of a uniformly convergent subsequence would follow from the Arzela lemma (cf. [4]). Indeed the demonstration of the uniform boundedness of | A'N\ is the key step in proving the existence of solutions to (1.1). In what follows it will be necessary to assume that the initial data satisfies the conditions 00 (2.5a) £/&<«>, y-i (2.5b) £/«/ < •• y-i This requirement on the initial data is, of course, stronger than the energy condition (1.6). Lemma 2.1. If the initial data (1.2) satisfies the condition (2.5), and oo (2.6) a0 + ax zZj «/ ^ 0, i-i there exists an interval 0^t<tc, such that \A'N\ is uniformly bounded independent of N on any closed subinterval 0^t^t*<te. Proof. After differentiating the function An, the Schwarz in- equality yields \An\ ^2a1J2f\T,,N\\T'i.N\ i=i (2.7) ^2ai(zfTl,N'tl\T'l.N)i\m Im i=i ' I °° «i in License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use 462 R. W. DICKEY [December Thus the object is to estimate the functions | TjN\. For this purpose define a function EitN as (cf. [5]) (2.8) EiJr = %^- + T),N ^ 0. 3 An The condition (2.6) guarantees that EjiN is defined in some neighbor- hood of t = 0 when N is sufficiently large (the condition (2.6) is equiva- lent to the requirement that the partial differential equation (1.4) be hyperbolic at 2 = 0). The functions Tj,n are solutions of (2.1); there- fore differentiation of (2.8) yields ,,,, j An((T'j,n)\^ \An'\ (2.9) EhN = - — ( . 1 ^—-Ej,N, An\ 3 An I An or equivalently a' J—!-\A'n\ drj \ . Estimates on both Tj,_yand T'jN follow from (2.10). Thus it is found that 2 2 2 2 / C ' I An I \ (2.11a) Tj,n Is (fij/j AN(0)+ «y) exp ( J ^—— dr j , (2.11b) (T'j,Ny/fANg «3)/fAN(0)+ a') exp (J ^—^- dr\ . Define Kn as 00 (2.12) Kn = Z (f$s/AN(0)+ fa)), 3=1 and note that finiteness of (2.12) follows from (2.5). In addition, the fact that AN(0) ^4N+i(0) shows that (2.13) P^+i S Kn- Combining (2.11b) and (2.7) it follows that (2.14) \AN\£2<OiAirKi,explj drJV , License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use 1969] NONLINEAR OSCILLATIONEQUATIONS 463 or d ( 1 r' I a'n\ \ (2.15)-exp (---L dr\ ^ (aiKNyi2 ^ (aiKM)112, dt \ 2 J o AN I if N^M. It is a consequence of (2.15) that / 1 r ' \ a'n I \ (2.16) exp— -!- dr)^ 1/(1 - (axKMY'H) for all A^ M and all t in the interval (2.17) 0^t<tM= l/(axKM)112. Combining (2.14) and (2.16), it is clear that when N^M and t is in the interval (2.17), | A'x\ satisfies the bound (2.18) \An\ ^2(axKM)lt2AN/(l - (axKM)ll2t). Since An is uniformly bounded independent of N, (2.18) shows that | A'N\ is also uniformly bounded independent of N in the interval (2.17). In fact, if K is defined as (2.19) A = lim Kn and tc is defined as (2.20) tc = 1/MO1'2 then \AN\ will be uniformly bounded in any closed interval O^t ^t*<tc. Q.E.D. It is of interest to note that, at least in the case where a0>0, the interval 0^t<tc grows arbitrarily large as the initial data (1.2) approaches zero. In view of the preceding remarks, Lemma 2.1 guarantees the exis- tence of a subsequence {^4^,} which converges uniformly to a (con- tinuous) function A(t) on any closed subinterval 0^t^t*<tc. Let Tj be the solution of the (linear) equation (2.21) T'/+j2A(t)Tj = 0, satisfying the initial conditions (1.2). There is no difficulty in showing that Tj,Ni—*Tj and T'jN.-*T'j on the interval 0^tf^t*<tc. The exis- tence of solutions to (1.1) is settled by the following Theorem 2.1. The infinite system of equations (1.1) have a solution satisfying the initial data (1.2) on any closed interval 0^t^t*<tcif the initial data satisfies the conditions (2.5) and (2.6). License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use 464 R. W. DICKEY [December Proof. It is only necessary to show that the solutions of the linear system (2.21) furnish a solution of the system (1.1). For this purpose it suffices to show that 00 (2.22) A(t) = ao + axT.fTl i=i The series which occurs in (2.22) converges since (cf. (2.11) and (2.16)) (2.23a) T\ = lim rjjr, ^ (ffi/l2A(0) + a\)/(i - axKu) t), (2.23b) (T[)2 = lim (PU.)2 ^ A(0i/A(O) + fa\)/(i - (aiKM) t), AT;-- for arbitrary M, and thus the series in (2.22) is majorized by a con- vergent series.